System and method for aircraft failure prediction

• US 10,457,423 B2
• Filed: 09/22/2017
• Issued: 10/29/2019
• Est. Priority Date: 09/22/2017
• Status: Active Grant

• Current Assignee: The Boeing Co.
• Original Assignee: The Boeing Co.
• Inventors: Lu, Tsai-Ching, Sundareswara, Rashmi, Huber, David, Waagen, Alex
• Primary Examiner(s): Sweeney, Brian P
• Application Number:
  US15/712,530
• Publication Number:
  US 20190092495A1
• Time in Patent Office: 767 Days
• Field of Search: 701 321
• CPC Class Codes: B64D 2045/0085 Devices for aircraft health monitoring, e.g. monitoring flutter or vibration
• CPC Class Codes: B64D 41/00 Power installations for auxiliary purposes
• CPC Class Codes: B64D 45/00 Aircraft indicators or protectors not otherwise provided for camouflage F41H3/00
• CPC Class Codes: B64F 5/40 Maintaining or repairing aircraft
• CPC Class Codes: B64F 5/60 Testing or inspecting aircraft components or systems
• CPC Class Codes: G06Q 10/06 Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
• CPC Class Codes: G06Q 10/20 Administration of product repair or maintenance
• CPC Class Codes: G07C 5/006 Indicating maintenance
• CPC Class Codes: G07C 5/0816 Indicating performance data, e.g. occurrence of a malfunction
• CPC Class Codes: G07C 5/085 using electronic data carriers

Abstract

An aircraft component failure prediction apparatus including a database and an aircraft maintenance controller coupled to the database. The controller classifies operational data, in a first classification, as corresponding with at least one maintenance message to form at least one classified multidimensional operation data matrix including classified data. The controller classifies the classified data of the at least one classified multidimensional data matrix, in a second classification, to predict an occurrence of a future maintenance message for the aircraft component. The controller preprocesses the at least one classified multidimensional operation data matrix, and generates an output of at least one partial dependency function that explains prediction of the occurrence of the future maintenance message.

Claims

1. An aircraft component failure prediction apparatus comprising:

a database coupled to at least one aircraft system, the database being configured to receive from the at least one aircraft system and store

at least one multidimensional operation data matrix corresponding to operation of an aircraft component obtained from a plurality of aircraft, and

at least one multidimensional maintenance message matrix corresponding to the operation of the aircraft component from the plurality of aircraft; and

an aircraft maintenance controller coupled to the database, the aircraft maintenance controller being configured to

classify operational data, in a first classification, of the at least one multidimensional operation data matrix as corresponding with at least one maintenance message to form at least one classified multidimensional operation data matrix including classified data;

classify the classified data of the at least one classified multidimensional operation data matrix with a machine learning model, in a second classification, to predict an occurrence of a future maintenance message for the aircraft component within a predetermined analysis time period, where the aircraft maintenance controller preprocesses the at least one classified multidimensional operation data matrix to remove classification obscuring data from the at least one classified multidimensional operation data matrix, and

generate an output of at least one partial dependency function on a user interface coupled to the aircraft maintenance controller, the output of the at least one partial dependency function being generated from the at least one classified multidimensional operation data matrix where the output of the at least one partial dependency function explains prediction of the occurrence of the future maintenance message by identifying at least which of the operational data is most frequently identified by the at least one maintenance message and operational ranges of the operational data most frequently identified by the at least one maintenance message.

2. The aircraft component failure prediction apparatus of claim 1, wherein the partial dependency function is a one dimensional partial dependency function whose output provides for visualizing trends for the classified data and ranges of values for the classified data that are indicative of the occurrence of the future maintenance message.

3. The aircraft component failure prediction apparatus of claim 1, wherein the partial dependency function is a two dimensional partial dependency function whose output provides for visualizing interactions between two dimensions of the classified data that are indicative of the occurrence of the future maintenance message.

4. The aircraft component failure prediction apparatus of claim 1, wherein the aircraft component is an auxiliary power unit.

5. The aircraft component failure prediction apparatus of claim 1, wherein the aircraft maintenance controller is further configured to:

obtain a collection of existing data, from the database, corresponding to a respective one of at least one maintenance message, where the collection of existing data corresponds to a plurality of aircraft and includes training data that comprises the existing data from all but one of the plurality of aircraft, and testing data that corresponds to the existing data of the one of the plurality of aircraft;

train the machine learning model with the training data so that the machine learning model indicates the future maintenance message will occur within the predetermined analysis time period; and

validate accuracy of the training of the machine learning model with the testing data.

6. The aircraft component failure prediction apparatus of claim 1, wherein the aircraft maintenance controller is configured to classify the operational data, in the first classification, by comparing a flight number in the at least one multidimensional maintenance message matrix for which a maintenance message exists with flight numbers in the at least one multidimensional operation data matrix to determine matching flight numbers for a predetermined maintenance message and whether the matching flight numbers are within a predetermined time period from each other.

7. The aircraft component failure prediction apparatus of claim 6, wherein the aircraft maintenance controller is configured to preprocess the at least one classified multidimensional operation data matrix by removing classified data from the at least one classified multidimensional operation data matrix so that the classified data remaining in the at least one classified multidimensional operation data matrix corresponds to a predetermined one of the aircraft and a predetermined one of the maintenance message for which the prediction of the occurrence of the future maintenance message of the aircraft component is to be made.

8. The aircraft component failure prediction apparatus of claim 1, wherein the aircraft maintenance controller is further configured to reduce a number of false positive failure indications in the second classification.

9. A method for predicting aircraft component failure, the method comprising:

receiving and storing, in a database coupled to at least one aircraft system, at least one multidimensional operation data matrix corresponding to operation of an aircraft component obtained from a plurality of aircraft, and at least one multidimensional maintenance message matrix corresponding to the operation of the aircraft component from the plurality of aircraft; and

with an aircraft maintenance controller coupled to the database

classifying operational data, in a first classification, of the at least one multidimensional operation data matrix as corresponding with at least one maintenance message to form at least one classified multidimensional operation data matrix including classified data;

classifying the classified data of the at least one classified multidimensional operation data matrix with a machine learning model, in a second classification, to predict an occurrence of a future maintenance message for the aircraft component within a predetermined analysis time period, where the at least one classified multidimensional operation data matrix is preprocessed to remove classification obscuring data from the at least one multidimensional operation data matrix, and

generating on a user interface coupled to the aircraft maintenance controller, a visualization that explains prediction of the occurrence of the future maintenance message by identifying at least which of the operational data is most frequently identified by the at least one maintenance message and operational ranges of the operational data most frequently identified by the at least one maintenance message.

10. The method of claim 9, further comprising, with the aircraft maintenance controller:

obtaining a collection of existing data, from the database, corresponding to a respective one of at least one maintenance message, where the collection of existing data corresponds to a plurality of aircraft and includes training data that comprises the existing data from all but one of the plurality of aircraft, and testing data that corresponds to the existing data of the one of the plurality of aircraft;

training the machine learning model with the training data so that the machine learning model indicates the future maintenance message will occur within the predetermined analysis time period; and

validating accuracy of the training of the machine learning model with the testing data.

11. The method of claim 9, further comprising classifying, with the aircraft maintenance controller, the operational data, in the first classification, by comparing a flight number in the at least one multidimensional maintenance message matrix for which a maintenance message exists with flight numbers in the at least one multidimensional operation data matrix to determine matching flight numbers for a predetermined maintenance message and whether the matching flight numbers are within a predetermined time period from each other.

12. The method of claim 11, further comprising, with the aircraft maintenance controller, preprocessing the at least one classified multidimensional operation data matrix by removing classified data from the at least one classified multidimensional operation data matrix so that the classified data remaining in the at least one classified multidimensional operation data matrix corresponds to a predetermined one of the aircraft and a predetermined one of the maintenance message for which the prediction of the occurrence of the future maintenance message of the aircraft component is to be made.

13. The method of claim 9, further comprising, with the aircraft maintenance controller, reducing a number of false positive failure indications in the second classification.

14. The method of claim 13, wherein reducing the number of false positive failure indications includes grouping flights into positive blocks that are indicative of the occurrence of the future maintenance message within the predetermined analysis time period, and into negative blocks that are indicative that the future maintenance message will not occur within the predetermined analysis time period.

15. The method of claim 14, further comprising:

grouping flights occurring within a predetermined blocking time period preceding a maintenance message occurrence into a positive block; and

grouping flights occurring after the maintenance message occurrence and flights occurring more than the predetermined blocking time period preceding the maintenance message occurrence into respective negative blocks.

16. The method of claim 15, further comprising identifying an entire positive block as being indicative of the occurrence of the future maintenance message when a predetermined number of flights within the entire positive block are classified as being indicative of the occurrence of the future maintenance message in the second classification.

17. The method of claim 16, further comprising dividing the respective negative blocks into sub-groups where a respective sub-group is identified as being a positive sub-group indicative of the occurrence of the future maintenance message when a predetermined number of flights within the respective sub-group are classified as being indicative of the occurrence of the future maintenance message in the second classification, otherwise the respective sub-group is identified as being a negative sub-group indicative that the future maintenance message will not occur within the predetermined analysis time period.

18. A method for predicting aircraft component failure, the method comprising:

receiving and storing, in a database coupled to at least one aircraft system, at least one multidimensional operation data matrix corresponding to operation of an aircraft component obtained from a plurality of aircraft, and at least one multidimensional maintenance message matrix corresponding to the operation of the aircraft component from the plurality of aircraft; and

with an aircraft maintenance controller coupled to the database

classifying operational data, in a first classification, of the at least one multidimensional operation data matrix as corresponding with at least one maintenance message to form at least one classified multidimensional operation data matrix including classified data;

classifying the classified data of the at least one classified multidimensional operation data matrix with a machine learning model, in a second classification, to predict an occurrence of a future maintenance message for the aircraft component within a predetermined analysis time period;

removing classified data from the at least one classified multidimensional operation data matrix so that the classified data remaining in the at least one classified multidimensional operation data matrix corresponds to a predetermined one of the aircraft and a predetermined one of the maintenance message for which the occurrence of the future maintenance message of the aircraft component is to be made, and

generating an output of at least one partial dependency function on a user interface coupled to the aircraft maintenance controller, the output of the at least one partial dependency function being generated from the at least one classified multidimensional operation data matrix where the output of the at least one partial dependency function explains prediction of the occurrence of the future maintenance message by identifying at least which of the operational data is most frequently identified by the at least one maintenance message and operational ranges of the operational data most frequently identified by the at least one maintenance message.

19. The method of claim 18, further comprising, with the aircraft maintenance controller:

obtaining a collection of existing data, from the database, corresponding to a respective one of at least one maintenance message, where the collection of existing data corresponds to a plurality of aircraft and includes training data that comprises the existing data from all but one of the plurality of aircraft, and testing data that corresponds to the existing data of the one of the plurality of aircraft;

training the machine learning model with the training data so that the machine learning model indicates the future maintenance message will occur within the predetermined analysis time period; and

validating accuracy of the training of the machine learning model with the testing data.

20. The method of claim 18, further comprising, with the aircraft maintenance controller, reducing a number of false positive failure indications in the second classification.